MEDIA PROCESSING & DELIVERY · FOR AI AGENTS
fal
Run hosted image, video and audio generation jobs.
Use fal’s MCP tools to discover models, inspect input schemas and pricing, submit generations and retrieve their completed outputs.
Your integration checklist
- Capabilities
- What you need from your owner
- Select a fal account or workspace, authorize the connection, fund the required credits and set a permitted model and spending scope.
- Authentication
- Current model-generation MCP uses OAuth at mcp.fal.ai/mcp-relay. Direct APIs and the separate genmedia CLI use API-key authentication.
- Your first task
- Search for a suitable model and inspect its schema and price before submitting one approved, bounded generation.
- Billing & payment
- Model APIs draw from prepaid credits with model-specific output units and prices. Inspect current pricing before each job; resolution, duration and model selection affect cost.
- Agent payment support
- Not verified. Programmatic usage does not establish autonomous checkout support.
- Operational limits
- Jobs can outlast a synchronous wait. Retain request IDs and retrieve existing results before submitting again. A prompt asking for cost approval does not create a server-enforced spending gate.
ORIGINAL AI AGENT STORE INTEGRATION PLAYBOOK
Treat generation as a tracked job with an output contract
Editorial recommendation · Reviewed 2026-10-15
When to choose this service
Choose fal when the task needs hosted generative-media models and inference capacity. Define the artifact’s format, dimensions or duration before choosing a model.
The dependency is trained model access and hosted inference hardware. Ordinary code can organize a media pipeline but does not replace its model weights and execution capacity.
Suggested workflow
- Search the catalog, inspect the selected model’s current input schema and calculate the expected cost within the task budget.
- Submit one bounded job and persist its endpoint and request ID immediately. Poll that job until it reaches a terminal state.
- Retrieve and save the completed asset, then check its format and content against the original output contract before publishing it.
Acceptance checks before real work
- Confirm the connected account and selected model before approving credit usage.
- Verify that a timeout resumes polling rather than submitting an identical second generation.
- Inspect the returned media itself; a successful job status alone does not prove it meets the brief.
If the workflow fails
If polling is interrupted, resume using the recorded request ID. If quality is inadequate, identify the failed requirement and approve a bounded revision instead of running an open-ended regeneration loop.